JP2006244289A - Image recognition device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image recognition device and an image recognition method for performing highly precise image recognition. <P>SOLUTION: A photograph image for a signal 10 taken by a camera 2 is projected to a principal component space formed by principal component analysis and a principal component coefficient for each principal component in the principal component space is calculated (S24). Using the principal component in which a cumulative contribution ratio found by adding up eigenvalues in the principal component in the descending order occupies a predetermined value, and a principal component coefficient of the principal component, an approximate image for the photograph image is formed (S26), and based on the approximate image, it is determined whether the signal in the photograph image shows a blue signal or not (S28). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image recognition apparatus and an image recognition method used for image recognition such as signs and traffic lights.

Conventionally, as an image recognition apparatus used for image recognition such as signs and traffic lights, as described in Japanese Patent Laid-Open No. 2000-293670, edge processing is performed on a photographed image, and signs are recognized based on the shape of the edge image. Things are known. This apparatus sets a slit on the photographed image, performs HSV conversion only on color data (RGB) relating to the slit, and determines whether or not there is a marker on the slit. As a result, the processing time required for label recognition is to be shortened.
JP 2000-293670 A

  In such a device, an edge appears even if it is not a sign, and it is necessary to remove these noises. It is difficult to efficiently remove the noise, and it is difficult to perform image recognition with high accuracy. Further, the photographed image is compared with the marker data prepared in advance to determine whether or not the photographed image is a predetermined marker image. However, it is necessary to compare images with a plurality of marker data, and the recognition process takes time. Take it. Furthermore, depending on the shooting environment, the accuracy of image comparison may be low, and the recognition accuracy may be low.

  Therefore, an object of the present invention is to provide an image recognition apparatus and an image recognition method that can perform image recognition with high accuracy.

  That is, an image recognition apparatus according to the present invention includes a photographing unit for photographing a recognition target object, a recording unit that records image data to be recognized, a principal component analyzing unit that performs principal component analysis on the image data, Projecting a captured image of the recognition object photographed by the photographing means onto a principal component space formed by principal component analysis by the principal component analyzing means, and calculating a principal component coefficient of each principal component in the principal component space. Approximation that forms an approximate image of the photographed image using a coefficient calculation means, a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value, and a principal component coefficient relating to the principal component The image forming unit includes a determination unit that determines whether the recognition target object of the captured image is a recognition target based on the approximate image.

  In the image recognition apparatus according to the present invention, the determination unit calculates a correlation value between the captured image and the approximate image, and determines whether the captured image is based on whether the correlation value exceeds a predetermined threshold value. It is determined whether or not the recognition target object is a recognition target.

  An image recognition apparatus according to the present invention includes a photographing unit for photographing a recognition target object, a recording unit that records image data to be recognized, a principal component analyzing unit that performs principal component analysis on the image data, An image extracting means for extracting a plurality of photographed images by shifting a position on the image of the recognition object photographed by the photographing means by a predetermined amount; and a main component formed by principal component analysis of the plurality of photographed images by the principal component analyzing means. Coefficient calculation means for projecting each into the component space and calculating a principal component coefficient of each principal component in the principal component space, and a principal component occupying a predetermined value with a cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one An approximate image forming unit that forms an approximate image for each of the plurality of captured images using a principal component coefficient related to the principal component, and a correlation value between the approximate image and the captured image, respectively. Out, it is configured to include a judgment means for the object recognition object of the captured image to determine whether a recognition object on the basis of the rate of change of the correlation value of the deviation amount of the image-extraction position.

  The image recognition apparatus according to the present invention records the captured image as image data to be recognized in the recording unit when the determination unit determines that the recognition target object of the captured image is a recognition target. It is characterized by that.

  The image recognition apparatus according to the present invention includes a photographing unit for photographing an object to be recognized, a recording unit that records image data of different recognition targets in the same category, and the image data of the different recognition targets, respectively. A principal component analysis means for analyzing, and a captured image of the recognition object photographed by the photographing means is projected onto each principal component space formed by principal component analysis by the principal component analysis means, and each principal component in the principal component space is projected Using the coefficient calculation means for calculating the principal component coefficient of the main component space, the principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value, and the principal component coefficient related to the principal component, Approximate image forming means for forming an approximate image of the captured image for each, and calculating a correlation value between the approximate image for each principal component space and the captured image, Zui and is configured to include a judgment means for judging whether the recognition target object of the captured image matches any recognition target.

  In the image recognition apparatus according to the present invention, the recognition target is a traffic light.

  The image recognition method according to the present invention includes a principal component analysis step for analyzing principal components of image data recorded in a recording means, and a captured image of a recognition object photographed by the photographing means as a principal component of the principal component analysis step. A coefficient calculation step of calculating the principal component coefficient of each principal component in the principal component space by projecting to the principal component space formed by the analysis, and a cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one has a predetermined value An approximate image forming step of forming an approximate image of the captured image using a main component occupied and a principal component coefficient related to the main component, and the recognition target object of the captured image based on the approximate image And a determination step for determining whether or not.

  In the image recognition method according to the present invention, in the determination step, a correlation value between the captured image and the approximate image is calculated, and the captured image is based on whether the correlation value exceeds a predetermined threshold value. It is judged whether the said to-be-recognized target object is a recognition target.

  The image recognition method according to the present invention includes a principal component analysis step of performing principal component analysis on the image data recorded in the recording means, and a plurality of positions on the image of the recognition object photographed by the photographing means by shifting by a predetermined amount. An image extraction step of extracting a photographed image, and the plurality of photographed images are respectively projected onto a principal component space formed by principal component analysis of the principal component analysis step, and a principal component coefficient of each principal component in the principal component space is calculated. A plurality of captured images are approximated by using a coefficient calculation step to calculate, a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value, and a principal component coefficient relating to the principal component. An approximate image forming step for forming an image, a correlation value between the approximate image and the captured image are calculated, and the imaging value is calculated based on a rate of change of the correlation value with respect to a shift amount of the image extraction position. The image is composed and a determination step of the recognition object is determined whether the recognition target.

  The image recognition method according to the present invention records the captured image as image data to be recognized in the recording means when it is determined in the determining step that the object to be recognized of the captured image is a recognition target. It is characterized by that.

  Further, the image recognition method according to the present invention includes a principal component analysis step for principal component analysis of image data of different recognition targets of the same category recorded in the recording means, and a photographed image of the recognition target object photographed by the photographing means. Is calculated on each principal component space formed by the principal component analysis of the principal component analysis step, and a coefficient calculation step for calculating a principal component coefficient of each principal component in the principal component space and a larger eigenvalue in the principal component An approximate image forming step of forming an approximate image of the photographed image for each principal component space using a principal component for which the cumulative contribution rate added from step occupies a predetermined value and a principal component coefficient relating to the principal component; and A correlation value between the approximate image for each principal component space and the captured image is calculated, and based on the correlation value, it is determined to which recognition target the recognized object of the captured image matches. It is constituted by a that decision steps.

  The image recognition method according to the present invention is characterized in that the recognition target is a traffic light.

  According to these inventions, the image data of the recognition object is subjected to principal component analysis, and the captured image of the recognition object is projected onto the principal component space formed by the principal component analysis to calculate the principal component coefficient of each principal component. Then, an approximate image of the photographed image is formed using the main component in which the cumulative contribution ratio obtained by adding the eigenvalues in the main component from the larger one occupies a predetermined value and the main component coefficient related to the main component. Since this approximate image is formed of principal components having a high cumulative contribution rate, the closer the recognition target object is to the recognition target, the closer to the captured image, and the recognition target object does not approximate the recognition target. It is far from the captured image. For this reason, it is possible to perform a recognition determination with high accuracy by determining whether or not the recognition target object of the captured image is a recognition target based on the approximate image. Further, since only one approximate image is required as a template image to be compared with the photographed image, the recognition process is simple and can be performed quickly in a short time.

  According to the present invention, it is possible to provide an image recognition apparatus and an image recognition method that can perform image recognition with high accuracy.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant description is omitted.
(First embodiment)
FIG. 1 is a schematic configuration diagram of an image recognition apparatus according to a first embodiment of the present invention.

  As shown in FIG. 1, an image recognition apparatus 1 according to the present embodiment is an apparatus that recognizes whether or not a photographed recognition target object is a recognition target, and includes a camera 2, an image processing unit 3, and a database 4. It is prepared for. The camera 2 is a photographing means for photographing the object to be recognized, and for example, a camera provided with an image sensor such as a CCD or C-MOS is used. The camera 2 is a camera capable of photographing a recognition target object as a color image.

  The image processing unit 3 is connected to the camera 2 and inputs a photographed image photographed by the camera 2, performs image processing on the photographed image, and determines whether or not a recognition target object of the photographed image is a recognition target. Image processing means. As the image processing unit 3, for example, an image memory, an image processor, and a signal output interface are used.

  The database 4 functions as a recording means for recording a plurality of image data to be recognized. The database 4 is connected to the image processing unit 3 and can input and output image data to and from the image processing unit 3. In the database 4, when the recognition target is a green signal, a plurality of image data of different green signals are recorded.

  Next, the operation of the image recognition apparatus 1 and the image recognition method according to the present embodiment will be described.

  FIG. 2 is a flowchart relating to principal component analysis processing in the image recognition apparatus 1 according to the present embodiment. FIG. 3 is a flowchart relating to recognition determination processing in the image recognition apparatus 1 according to the present embodiment, and illustrates an image recognition method according to the present embodiment. 2 and 3 is executed by the image processing unit 3, for example.

  Based on FIG. 2, the principal component analysis process will be described.

  First, as principal component analysis processing, image data is read (S10). This reading of image data is a process of reading a plurality of recognition target image data stored in the database 4 into the image processing unit 3. Then, after reading this image data, a principal component analysis process is performed on the image data to be recognized (S12). The principal component analysis process is a process for forming a principal component space by performing principal component analysis on image data to be recognized and deriving a base image.

  For example, as shown in FIG. 4, based on image data, base images g1, g2, g3, g4, g5... Gn and an average image g0 are derived. The base images g1, g2,... Gn are arranged in descending order of eigenvalues. The average image g0 is an image obtained by averaging the image elements of each piece of image data to be recognized. A principal component space is formed based on these average image g0, base images g1, g2, g3, g4, g5. At that time, each of the base images g1, g2, g3, g4, g5... Gn constitutes an axis that determines the number of dimensions of the principal component space. When n base images exist, the number of dimensions of the principal component space is n.

  Next, the recognition determination process will be described with reference to FIG.

  First, as a recognition determination process, a captured image is read (S20). The captured image reading process is a process of taking a captured image captured by the camera 2 into the image processing unit 3. For example, as shown in FIG. 1, a photographed image obtained by photographing the traffic light 10 that is the recognition target object with the camera 2 is taken into the image processing unit 3 as image data. Then, an image extraction process is performed (S22). The image extraction process is a process for extracting a shooting area of the traffic light 10 from the shot image. By this processing, the extracted shooting area is set as a shot image of the recognition target object.

  Then, a principal component coefficient calculation process is performed (S24). The principal component coefficient calculation process is a process of projecting the captured image of the recognition target object extracted in S22 onto the principal component space and calculating the principal component coefficients of the respective principal components. For example, the principal component coefficients c1, c2,... Cn of each principal component are calculated by performing KL transformation using the eigenvectors and eigenvalues of the base image as the principal components as parameters.

  Then, the process proceeds to S26, and an approximate image forming process is performed. The approximate image forming process is a process for forming an approximate image of a captured image using the average image g0, a predetermined base image, and a predetermined principal component coefficient. That is, an approximate image of a photographed image is formed by using a main component in which the cumulative contribution ratio obtained by adding the eigenvalues in the main component from the larger one occupies a predetermined value and a main component coefficient related to the main component. For example, when the predetermined value of the cumulative contribution rate is 90% and the base images that are the main components occupying it are g1 to g5, the approximate image S of the photographed image is expressed by the following equation (1).

S = g0 + c1 · g1 + c2 · g2 + c3 · g3 + c4 · g4 + c5 · g5 (1)
This approximate image S is obtained by restoring the captured image based on the average image g0 and the base images g1 to 5, and information on the main components of the portion exceeding the cumulative contribution rate of 90%, that is, the base images g6, g7. The image has been removed. In the approximate image forming process, the predetermined value of the cumulative contribution rate is preferably set to 80 to 95%. Thus, by setting the cumulative contribution rate to 80 to 95%, it is possible to reduce the calculation load without significantly reducing the recognition rate.

  Then, the process proceeds to S28, and recognition determination processing is performed. The recognition determination process is a process of determining whether or not the traffic light 10 that is the recognition target object of the captured image is a green signal that is a recognition target based on the approximate image. This recognition determination process is performed, for example, by using a normalized correlation or difference summation method.

  5 to 7 show examples of the recognition determination process.

  FIG. 5 shows a correlation value in a case where a recognition determination is made on a captured image obtained by capturing a green traffic light as an object to be recognized. As shown in FIGS. 5A to 5E, since the recognition target object is a green signal that is a recognition target, the approximate image approximates the captured image (original image), and the correlation value is also It is a very high value of 9692-9896.

  FIG. 6 shows a case where a recognized image is taken for a signal other than a green signal as an object to be recognized (FIGS. 6A and 6B are yellow traffic lights and FIGS. 6C to 6E are red traffic lights). The correlation value is shown. As shown in FIGS. 6A to 6E, since the recognition target objects are the yellow traffic light and the red traffic light that are not the recognition target, the approximate image is not so close to the captured image, and the correlation The value is slightly lower, 6314-8563.

  FIG. 7 shows a correlation value when a recognition determination is performed on a captured image obtained by capturing an object other than a traffic light as a recognition target object. As shown in FIGS. 7A to 7E, since the object to be recognized is something other than a traffic light that is not a recognition object, the approximate image is not so close to the captured image, and the correlation value Is slightly lower, 4195-7236.

  In the recognition determination in FIGS. 5 to 7, for example, a correlation value 9000 is set as a recognition determination threshold value, and a correlation value exceeding the threshold value is determined as a green traffic light, and a correlation equal to or lower than the threshold value is determined. By determining that the value is not a green traffic light, an accurate recognition determination can be made.

  As described above, according to the image recognition apparatus and the image recognition method according to the present embodiment, the principal component analysis is performed on the image data of the green signal that is the recognition target, and the recognition target is formed in the principal component space formed by the principal component analysis. Calculate the principal component coefficient of each principal component by projecting a photographed image of the object, and add the eigenvalue of the principal component from the larger one, and the principal component that occupies the predetermined value and the principal component coefficient related to that principal component Is used to form an approximate image of the captured image. Since this approximate image is formed of principal components having a high cumulative contribution rate, the closer the recognition target object is to the recognition target, the closer to the captured image, and the recognition target object does not approximate the recognition target. It is far from the captured image. For this reason, by determining whether or not the recognition target object of the photographed image is a recognition target based on the approximate image, it is possible to perform a highly accurate recognition determination with few erroneous recognitions and recognition misses.

  Further, according to the image recognition apparatus and the image recognition method according to the present embodiment, the template image to be compared with the photographed image is only one approximate image, so that the recognition process is simple and can be performed in a short time. That is, the recognition processing can be performed quickly in a short time compared to the conventional pattern matching in which recognition processing is performed by comparing a plurality of templates and captured images.

  In the present embodiment, the case where image data that is a color image in the RGB color space is directly recognized and determined by principal component analysis has been described. However, the image recognition apparatus and the image recognition method according to the present invention are as such. The present invention is not limited to this, and image data may be subjected to color conversion, and the color-converted image may be recognized and determined by principal component analysis.

  For example, a plurality of image data recorded in the database 4 is converted from an RGB color space to an HSI color space, and each color space (H color space, S color space, I color space) converted into the HSI color space is mainly used. Perform component analysis. Then, each captured image is projected onto the principal component space for each color space, the principal component coefficient of each principal component in the principal component space is calculated, and the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one is a predetermined value. An approximate image of the photographed image is formed using the main component that occupies and the main component coefficient related to the main component. Then, as described above, it is determined whether or not the recognition target object of the photographed image is a recognition target based on the approximate image. Note that the number of dimensions of the principal component to be removed by the cumulative contribution rate may be an average of the number of dimensions obtained in each color space.

  By performing color conversion in this way and performing image recognition processing, recognition determination with higher accuracy can be performed as compared with the above-described image recognition.

In the present embodiment, the case where the recognition target is a green traffic light has been described. However, the image recognition apparatus and the image recognition method according to the present invention are not limited to such a case, and the recognition target is a red traffic light, a yellow traffic light. It may be in the case of traffic lights, road signs, vehicles or others.
(Second embodiment)
Next, an image recognition apparatus and an image recognition method according to the second embodiment will be described.

  The image recognition apparatus according to the present embodiment has a hardware configuration similar to that of the image recognition apparatus 1 according to the first embodiment shown in FIG. 1, and performs control processes substantially similar to those in FIGS. However, different image areas are set for the images captured in the image extraction process of S22 of the recognition determination control of FIG. 3, and the correlation values of the recognition determination are obtained by performing the processes of S24 to 28 on the captured images of these areas. It is calculated, and it is determined whether or not the recognition target object of the captured image is a recognition target based on the change rate of each correlation value.

  For example, as illustrated in FIG. 8, a captured image 21 is set by shifting the extraction position by a predetermined amount with respect to a captured image 20 extracted by extracting a portion where the traffic light 10 is captured as a captured image. Then, each of the captured images 20 and 21 is subjected to the processing of S24 to S28 in FIG. 3 to calculate each correlation value. Then, based on the rate of change of the correlation value with respect to the displacement amount of the extraction position, it is determined whether or not the traffic light 10 is a green signal that is a recognition target.

  9 and 10 show examples of recognition determination using the image recognition method according to the present embodiment.

  FIG. 9 shows a correlation value calculated by shifting the extraction position by a predetermined amount in the X-axis direction and the Y-axis direction for a captured image obtained by capturing a green traffic light as a recognition object. FIG. 10 shows a correlation value calculated by shifting the extraction position by a predetermined amount in the X-axis direction and the Y-axis direction for a captured image obtained by capturing a pedestrian's message as a recognition object. In the correlation value graphs of FIGS. 9 and 10, the vertical axis is the correlation value, and the horizontal axis is the amount of deviation of the extraction position.

  When the correlation value graph of FIG. 9 is seen, the correlation value has a peak value at the center position, and the correlation value gradually changes even when the extraction position is shifted from the peak value. It can be seen that the change rate of the correlation value is small.

  On the other hand, in the correlation value graph of FIG. 10, the peak of the correlation value is not clear, the correlation value changes rapidly by shifting the extraction position, and the rate of change of the correlation value with respect to the amount of deviation of the extraction position is large. I understand.

For this reason, for example, when the rate of change of the correlation value with respect to the shift amount of the extracted position of the captured image is equal to or greater than a predetermined change rate, it is determined that the recognition target is not a recognition target, When the change rate of the correlation value with respect to is not equal to or higher than the predetermined change rate, it is possible to make a highly accurate image recognition determination by determining that the recognition target object is a recognition target. In particular, if recognition determination is performed based on the correlation value change rate with respect to the correlation value peak value and the amount of deviation of the extracted position of the captured image, more accurate image recognition determination can be performed.
(Third embodiment)
Next, an image recognition apparatus and an image recognition method according to the third embodiment will be described.

  The image recognition apparatus according to the present embodiment has a hardware configuration similar to that of the image recognition apparatus 1 according to the first embodiment shown in FIG. 1, and performs control processes substantially similar to those in FIGS. However, when it is determined in the recognition determination process in S22 of FIG. 3 that the recognition target object of the captured image is a recognition target, the captured image is recorded in the database 4 as image data to be recognized. In the recognition determination process, when it is determined that the recognition target object of the captured image is not a recognition target, the captured image is preferably recorded separately in the database 4 as other types of image data.

  FIG. 11 shows a flowchart of the operation of the image recognition apparatus and the image recognition method according to this embodiment. In the control process of the flowchart of FIG. 11, first, a captured image is read in S40. Then, image extraction processing, principal component coefficient calculation processing, approximate image formation processing, and recognition determination processing are sequentially performed (S42, S44, S46, S48). The processing of S40 to 48 is performed in the same manner as the processing of S20 to 28 in FIG.

  Then, the process proceeds to S50, and it is determined whether or not the recognition target object of the captured image is determined to be a recognition target in S48. For example, it is determined whether or not the traffic light 11 of the captured image is determined to be a green signal. When it is determined that the recognition target object of the captured image is the recognition target, the captured image is recorded in the database 4 as image data to be recognized (S52). Thereby, the image data to be recognized can be increased.

  On the other hand, when it is determined that the recognition target object of the captured image is not a recognition target, the captured image is recorded in the database 4 as other types of image data (S54). For example, when the traffic light 11 of the captured image is a red signal, the captured image is recorded in the database 4 as image data other than the blue signal.

As described above, according to the image recognition apparatus and the image recognition method according to the present embodiment, a highly accurate recognition determination can be performed as in the image recognition apparatus and the image recognition method according to the first embodiment, and for each recognition determination. Image data can be accumulated in the beta base 4. For this reason, the accumulated data in the database 4 can be enriched to enable more accurate recognition judgment.
(Fourth embodiment)
Next, an image recognition apparatus and an image recognition method according to the fourth embodiment will be described.

  In the first embodiment to the third embodiment described above, the case where there is one recognition target (for example, the case where the recognition target is only a green signal) has been described, but the image recognition apparatus and the image recognition method according to the present embodiment are different. The recognition target of the captured image is set for each of the plurality of recognition targets, the closest recognition target is determined, and the recognition target object of the captured image is the closest recognition target. to decide.

  For example, when the traffic light is an object to be recognized, the blue signal image data, the red signal image data, and the yellow signal image data are recorded in the database 4 respectively, and each of them is the main component as shown in S12 of FIG. Analysis processing is performed to form a principal component space for each signal. Then, as shown in S20 and S22 of FIG. 3, a captured image reading process and an image extracting process are performed. Then, for the captured image, the blue signal, the red signal, and the yellow signal are projected onto the principal component space, the principal component coefficient is calculated (S24), the approximate image is formed (S26), and the correlation value is calculated by the recognition determination process (S28). Do each. Then, the correlation values of the blue signal, the red signal, and the yellow signal are respectively compared to determine the one with the highest correlation value, and it is determined that the traffic signal of the captured image is the signal with the highest correlation value. At that time, if the correlation values of the blue signal, the red signal, and the yellow signal are all lower than the predetermined value, it is determined that the recognition target object is something other than the traffic light.

  Thus, according to the image recognition device and the image recognition method according to the present embodiment, it is possible to identify different types of recognition targets in the same category. Further, when the correlation values of the recognition targets in the same category are all lower than the predetermined value, it is possible to determine that the recognition target object is other than the same category and exclude the different categories.

  In the image recognition apparatus and the image recognition method according to the present embodiment, similarity range values of different types of recognized objects of the same category are calculated, and the recognized objects are of a predetermined type of the same category. It may be recognized whether or not the object to be recognized belongs to the same category.

  That is, when various principal component spaces of the same category are formed, different types of image data of the same category are projected onto the principal component space, an approximate image is formed, a recognition judgment process is performed, and a correlation value is calculated. deep. This correlation value is recorded as the similarity range value of the same category of the type. If the recognition value of the recognition target object is recognized and the correlation value is lower than the similarity range value, it can be determined that the recognition target object is of a different category.

  For example, image data is prepared for each of a blue signal, a red signal, and a yellow signal of a traffic light, and a principal component space is formed for each. Then, the image data of the red signal and the yellow signal are respectively projected onto the principal component space of the blue signal, an approximate image is formed, and recognition determination processing is performed to calculate a correlation value. This correlation value is recorded as the similarity range value of the traffic light. When the recognition target object is recognized and determined, if the correlation value is lower than the similarity range value, it can be determined that the recognition target object is something other than a traffic light.

  In the present embodiment, the case where a traffic signal is a recognition target has been described, but the present invention may be applied to other categories. For example, using a vehicle or a sign as a recognition target, it may be determined whether the recognition target object is a vehicle or whether the recognition target object is a sign.

1 is a schematic configuration diagram of an image recognition device according to a first embodiment of the present invention. It is a flowchart which shows the principal component analysis process in the image recognition apparatus of FIG. It is a flowchart which shows operation | movement of the image recognition apparatus and image recognition method which concern on 1st embodiment. It is explanatory drawing of the Example of the recognition judgment process in the image recognition apparatus and image recognition method which concern on 1st embodiment. It is explanatory drawing of the Example of the recognition judgment process in the image recognition apparatus and image recognition method which concern on 1st embodiment. It is explanatory drawing of the Example of the recognition judgment process in the image recognition apparatus and image recognition method which concern on 1st embodiment. It is explanatory drawing of the Example of the recognition judgment process in the image recognition apparatus and image recognition method which concern on 1st embodiment. It is explanatory drawing of the image recognition apparatus and image recognition method which concern on 2nd embodiment. It is explanatory drawing of the image recognition apparatus and image recognition method which concern on 2nd embodiment. It is explanatory drawing of the image recognition apparatus and image recognition method which concern on 2nd embodiment. It is a flowchart which shows the operation | movement of the image recognition apparatus and image recognition method which concern on 3rd embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Image recognition apparatus, 2 ... Camera (photographing means), 3 ... Image processing part, 4 ... Database (recording means), 10 ... Traffic light.

Claims (12)

Photographing means for photographing the object to be recognized;
A recording means for recording image data to be recognized;
A principal component analysis means for principal component analysis of the image data;
Projecting a captured image of the recognition object photographed by the photographing means onto a principal component space formed by principal component analysis by the principal component analyzing means, and calculating a principal component coefficient of each principal component in the principal component space. Coefficient calculation means;
An approximate image forming means for forming an approximate image of the photographed image using a main component occupying a predetermined value with a cumulative contribution ratio obtained by adding eigenvalues in the main component from a larger one and a main component coefficient related to the main component;
Determination means for determining whether or not the recognition target object of the captured image is a recognition target based on the approximate image;
An image recognition apparatus comprising: The determination unit calculates a correlation value between the captured image and the approximate image, and the recognition target object of the captured image is a recognition target based on whether the correlation value exceeds a predetermined threshold value. To determine whether or not
The image recognition apparatus according to claim 1. Photographing means for photographing the object to be recognized;
A recording means for recording image data to be recognized;
A principal component analysis means for principal component analysis of the image data;
Image extracting means for extracting a plurality of photographed images by shifting the position on the image of the recognition object photographed by the photographing means by a predetermined amount;
Coefficient calculation means for projecting each of the plurality of captured images onto a principal component space formed by principal component analysis by the principal component analysis means, and calculating a principal component coefficient of each principal component in the principal component space;
Approximate image formation that forms an approximate image for each of the plurality of captured images using a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value and a principal component coefficient related to the principal component Means,
A correlation value between the approximate image and the photographed image is calculated, and it is determined whether or not the recognition target object of the photographed image is a recognition target based on a change rate of the correlation value with respect to the shift amount of the image extraction position. A judgment means to
An image recognition apparatus comprising:   4. When the determination unit determines that the recognition target object of the captured image is a recognition target, the captured image is recorded as recognition target image data in the recording unit. The image recognition apparatus in any one of. Photographing means for photographing the object to be recognized;
Recording means for recording different recognition target image data of the same category;
Principal component analysis means for principal component analysis of each of the different recognition target image data,
A coefficient for calculating a principal component coefficient of each principal component in the principal component space by projecting a captured image of the recognition object photographed by the photographing means onto each principal component space formed by principal component analysis by the principal component analyzing means. A calculation means;
An approximate image of the photographed image is formed for each principal component space using a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value and a principal component coefficient related to the principal component. An approximate image forming means;
Determining means for calculating a correlation value between the approximate image for each principal component space and the captured image, and determining to which recognition object the recognition target object of the captured image matches based on the correlation value; ,
An image recognition apparatus comprising:   The image recognition apparatus according to claim 1, wherein the recognition target is a traffic light. A principal component analysis step for principal component analysis of the image data recorded in the recording means;
Coefficient calculation for projecting a captured image of the recognition object photographed by the photographing means onto a principal component space formed by principal component analysis of the principal component analysis step, and calculating principal component coefficients of each principal component in the principal component space Process,
An approximate image forming step of forming an approximate image of the photographed image using a main component occupying a predetermined value with a cumulative contribution ratio obtained by adding eigenvalues from the larger one in the main component and a main component coefficient related to the main component;
A determination step of determining whether or not the recognition target object of the captured image is a recognition target based on the approximate image;
An image recognition method comprising: In the determination step, a correlation value between the captured image and the approximate image is calculated, and the recognition target object of the captured image is a recognition target based on whether the correlation value exceeds a predetermined threshold value. To determine if there is,
The image recognition method according to claim 7. A principal component analysis step for principal component analysis of the image data recorded in the recording means;
An image extraction step of extracting a plurality of photographed images by shifting a position on the image of the recognition object photographed by the photographing means by a predetermined amount;
A coefficient calculation step of projecting each of the plurality of captured images onto a principal component space formed by principal component analysis of the principal component analysis step, and calculating a principal component coefficient of each principal component in the principal component space;
Approximate image formation that forms an approximate image for each of the plurality of captured images using a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value and a principal component coefficient related to the principal component Process,
Correlation values between the approximate image and the captured image are calculated, respectively, and it is determined whether or not the recognition target object of the captured image is a recognition target based on a rate of change of the correlation value with respect to the shift amount of the image extraction position. A decision process to
An image recognition method comprising:   10. When the determination step determines that the recognition target object of the captured image is a recognition target, the captured image is recorded as recognition target image data in the recording unit. An image recognition method according to any one of the above. A principal component analysis step for principal component analysis of image data of different recognition targets of the same category recorded in the recording means,
The captured image of the object to be recognized photographed by the photographing means is projected onto each principal component space formed by principal component analysis in the principal component analysis step, and the principal component coefficients of each principal component in the principal component space are calculated. Coefficient calculation step;
An approximate image of the photographed image is formed for each principal component space using a principal component in which the cumulative contribution ratio obtained by adding the eigenvalues in the principal component from the larger one occupies a predetermined value and a principal component coefficient related to the principal component. An approximate image forming step,
A determination step of calculating a correlation value between the approximate image for each principal component space and the captured image, and determining to which recognition object the recognition target object of the captured image matches based on the correlation value; ,
An image recognition method comprising: The image recognition method according to claim 7, wherein the recognition target is a traffic light.

Images